Wall Street Now Treats AI Infrastructure as Bonds, Not Tech

The financial sector has fundamentally reclassified how it views AI infrastructure. When JPMorgan projects $30 billion to $40 billion in annual data center securitization and Morgan Stanley forecasts up to $300 billion in hyperscaler debt issuance for 2026 alone, the message is clear: Wall Street now treats data centers as infrastructure bonds, not technology capex.

Goldman Sachs estimates cumulative AI infrastructure capital expenditure will reach approximately $7.6 trillion from 2026 through 2031, covering compute hardware, data centers, and power systems. The five largest hyperscalers—Amazon, Microsoft, Alphabet, Meta, and Oracle—will collectively spend between $90 billion and $100 billion in 2026, nearly double their 2025 investment.

Why it matters

This asset-class shift exposes a structural mismatch: capital markets have unified fiber, electricity, and compute into a single investment category, while federal agencies still regulate them through separate silos with separate timelines. That gap creates friction precisely when AI deployment demands coordinated infrastructure at unprecedented speed and scale.

Power becomes the binding constraint

The limiting factor for AI infrastructure is no longer silicon or software—it's electricity. A single AI training facility requires 100 to 500 megawatts of continuous power, equivalent to a small city's demand. Goldman Sachs projects global data center power consumption will surge 220 percent from 2023 levels to reach 1,350 terawatt-hours by 2030. Gartner estimates that power shortages could constrain 40 percent of AI data centers by 2027.

The Federal Energy Regulatory Commission approved significant grid interconnection reforms on June 18, targeting a reduction in connection timelines from multi-year backlogs to roughly 90 days. But FERC's authority stops at queue management. It cannot accelerate transmission permitting, site new generation capacity, or resolve the multi-state coordination required to move electricity from production to consumption points.

Hyperscalers build around shared infrastructure

Facing reliability constraints in commodity infrastructure, the largest cloud providers have made a rational business decision: build private alternatives. Amazon, Microsoft, and Google are constructing dedicated fiber networks, signing behind-the-meter power agreements, and designing proprietary cooling systems. This strategy makes business sense when revenue depends on reliability guarantees that shared infrastructure cannot provide at AI scale.

The policy challenge emerges when the largest potential customers opt out of shared systems. Traditional carriers and utilities operate under must-serve obligations and rate regulation. Hyperscalers invest or withdraw through private contracts without comparable public accountability. This asymmetric regulatory structure creates economic pressure on shared infrastructure precisely when coordination matters most.

The equity dimension

Connectivity quality determines which households can participate in AI's economic benefits. Fiber connections enable heavier AI workloads than DSL or fixed wireless access. As AI infrastructure concentrates in corridors with adequate power and connectivity, regions without that foundation risk exclusion from economic gains.

Reform opportunities exist without creating new agencies or subsidy programs. Streamlining National Environmental Policy Act review processes for grid infrastructure—particularly transmission lines and interconnections—would address the most time-consuming bottleneck in the energy stack. Transmission permitting reform has bipartisan support in principle but has repeatedly failed when bundled with unrelated legislative priorities.

These details were first reported by Shane Tews at the American Enterprise Institute.